import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
from sklearn import datasets

from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error, r2_score
from sklearn.model_selection import train_test_split

df = pd.read_csv('beer_profile_and_ratings.csv')
df.head()

	Name	Style	Brewery	Beer Name (Full)	Description	ABV	Min IBU	Max IBU	Astringency	Body	...	Fruits	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews
0	Amber	Altbier	Alaskan Brewing Co.	Alaskan Brewing Co. Alaskan Amber	Notes:Richly malty and long on the palate, wit...	5.3	25	50	13	32	...	33	57	8	111	3.498994	3.636821	3.556338	3.643863	3.847082	497
1	Double Bag	Altbier	Long Trail Brewing Co.	Long Trail Brewing Co. Double Bag	Notes:This malty, full-bodied double alt is al...	7.2	25	50	12	57	...	24	35	12	84	3.798337	3.846154	3.904366	4.024948	4.034304	481
2	Long Trail Ale	Altbier	Long Trail Brewing Co.	Long Trail Brewing Co. Long Trail Ale	Notes:Long Trail Ale is a full-bodied amber al...	5.0	25	50	14	37	...	10	54	4	62	3.409814	3.667109	3.600796	3.631300	3.830239	377
3	Doppelsticke	Altbier	Uerige Obergärige Hausbrauerei GmbH / Zum Uerige	Uerige Obergärige Hausbrauerei GmbH / Zum Ueri...	Notes:	8.5	25	50	13	55	...	49	40	16	119	4.148098	4.033967	4.150815	4.205163	4.005435	368
4	Sleigh'r Dark Doüble Alt Ale	Altbier	Ninkasi Brewing Company	Ninkasi Brewing Company Sleigh'r Dark Doüble A...	Notes:Called 'Dark Double Alt' on the label.Se...	7.2	25	50	25	51	...	11	51	20	95	3.625000	3.973958	3.734375	3.765625	3.817708	96

5 rows × 25 columns

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 3197 entries, 0 to 3196
Data columns (total 25 columns):
 #   Column             Non-Null Count  Dtype  
---  ------             --------------  -----  
 0   Name               3197 non-null   object 
 1   Style              3197 non-null   object 
 2   Brewery            3197 non-null   object 
 3   Beer Name (Full)   3197 non-null   object 
 4   Description        3197 non-null   object 
 5   ABV                3197 non-null   float64
 6   Min IBU            3197 non-null   int64  
 7   Max IBU            3197 non-null   int64  
 8   Astringency        3197 non-null   int64  
 9   Body               3197 non-null   int64  
 10  Alcohol            3197 non-null   int64  
 11  Bitter             3197 non-null   int64  
 12  Sweet              3197 non-null   int64  
 13  Sour               3197 non-null   int64  
 14  Salty              3197 non-null   int64  
 15  Fruits             3197 non-null   int64  
 16  Hoppy              3197 non-null   int64  
 17  Spices             3197 non-null   int64  
 18  Malty              3197 non-null   int64  
 19  review_aroma       3197 non-null   float64
 20  review_appearance  3197 non-null   float64
 21  review_palate      3197 non-null   float64
 22  review_taste       3197 non-null   float64
 23  review_overall     3197 non-null   float64
 24  number_of_reviews  3197 non-null   int64  
dtypes: float64(6), int64(14), object(5)
memory usage: 624.5+ KB

df.describe()

	ABV	Min IBU	Max IBU	Astringency	Body	Alcohol	Bitter	Sweet	Sour	Salty	Fruits	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews
count	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000	3197.000000
mean	6.526688	21.180482	38.986863	16.515796	46.129496	17.055990	36.364404	58.270879	33.145449	1.017204	38.529559	40.924617	18.345637	75.330935	3.638789	3.754393	3.660428	3.702496	3.747522	233.284955
std	2.546997	13.242242	21.355281	10.410661	25.947842	17.331334	25.791152	34.281310	35.780172	2.132651	32.296646	30.403641	23.756582	39.909338	0.503209	0.403416	0.449937	0.510361	0.444288	361.811847
min	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	1.509615	1.571429	1.285714	1.214286	1.136364	1.000000
25%	5.000000	15.000000	25.000000	9.000000	29.000000	6.000000	17.000000	33.000000	11.000000	0.000000	12.000000	18.000000	4.000000	45.000000	3.422559	3.604651	3.470021	3.500000	3.566667	23.000000
50%	6.000000	20.000000	35.000000	14.000000	40.000000	11.000000	31.000000	54.000000	22.000000	0.000000	29.000000	33.000000	10.000000	73.000000	3.720183	3.833333	3.741667	3.791667	3.830239	93.000000
75%	7.600000	25.000000	45.000000	21.000000	58.000000	22.000000	52.000000	77.000000	42.000000	1.000000	60.000000	56.000000	23.000000	103.000000	3.978000	4.000000	3.965587	4.033333	4.032847	284.000000
max	57.500000	65.000000	100.000000	81.000000	175.000000	139.000000	150.000000	263.000000	284.000000	48.000000	175.000000	172.000000	184.000000	239.000000	5.000000	4.666667	5.000000	5.000000	5.000000	3290.000000

## Assess correlations visually:
sns.pairplot(df)

<seaborn.axisgrid.PairGrid at 0x7fa7ebaa06a0>

# I think all of the review types are averaged into 'review_overall' rating, so I tried to spot other (not as intuitive) correlations.
df.plot.scatter(x='Body',y='Malty')
df.plot.scatter(x='Bitter', y='Hoppy')
df.plot.scatter(x='Sour', y='Astringency')

<AxesSubplot:xlabel='Sour', ylabel='Astringency'>

##Check for correlations numerically:
df.corr()

	ABV	Min IBU	Max IBU	Astringency	Body	Alcohol	Bitter	Sweet	Sour	Salty	Fruits	Hoppy	Spices	Malty	review_aroma	review_appearance	review_palate	review_taste	review_overall	number_of_reviews
ABV	1.000000	0.432005	0.501037	-0.169521	0.241673	0.654908	0.067388	0.463487	0.100795	-0.120089	0.291001	-0.052596	0.191468	0.162060	0.416197	0.349139	0.392517	0.381200	0.251838	0.142176
Min IBU	0.432005	1.000000	0.854248	-0.071501	0.325338	0.323694	0.539452	0.227139	-0.073098	-0.057512	0.066335	0.407475	-0.046152	0.300041	0.343730	0.380908	0.352723	0.339885	0.289972	0.219822
Max IBU	0.501037	0.854248	1.000000	-0.120273	0.310617	0.392811	0.478080	0.277292	-0.043275	-0.083214	0.172929	0.345168	0.044533	0.288219	0.369456	0.379655	0.355608	0.346557	0.271969	0.221572
Astringency	-0.169521	-0.071501	-0.120273	1.000000	-0.059540	-0.171987	0.114686	-0.021456	0.571030	0.347155	0.345232	0.330951	-0.083795	-0.082085	0.068579	0.070948	0.106950	0.092021	0.159788	-0.050062
Body	0.241673	0.325338	0.310617	-0.059540	1.000000	0.268885	0.542236	0.458842	-0.126733	-0.099277	-0.048155	0.070138	0.185123	0.754228	0.360234	0.421331	0.345270	0.365055	0.313824	0.039057
Alcohol	0.654908	0.323694	0.392811	-0.171987	0.268885	1.000000	0.009088	0.527039	0.048767	-0.094329	0.254299	-0.079949	0.252876	0.270106	0.242989	0.184693	0.205569	0.206701	0.069789	0.048346
Bitter	0.067388	0.539452	0.478080	0.114686	0.542236	0.009088	1.000000	0.091705	-0.136914	0.004693	-0.093450	0.712887	-0.084048	0.565570	0.241457	0.331854	0.257267	0.253373	0.260874	0.112026
Sweet	0.463487	0.227139	0.277292	-0.021456	0.458842	0.527039	0.091705	1.000000	0.257913	-0.131918	0.482030	-0.034327	0.107548	0.471032	0.356514	0.332311	0.315767	0.322277	0.207293	0.048487
Sour	0.100795	-0.073098	-0.043275	0.571030	-0.126733	0.048767	-0.136914	0.257913	1.000000	0.098173	0.785883	0.068895	0.001831	-0.303266	0.289320	0.204817	0.245169	0.244314	0.213449	0.006642
Salty	-0.120089	-0.057512	-0.083214	0.347155	-0.099277	-0.094329	0.004693	-0.131918	0.098173	1.000000	0.026920	0.172606	-0.023079	-0.028241	-0.088210	-0.081388	-0.038413	-0.057479	-0.008199	-0.029680
Fruits	0.291001	0.066335	0.172929	0.345232	-0.048155	0.254299	-0.093450	0.482030	0.785883	0.026920	1.000000	0.110407	0.148281	-0.196890	0.386676	0.288594	0.324906	0.319690	0.261504	0.106223
Hoppy	-0.052596	0.407475	0.345168	0.330951	0.070138	-0.079949	0.712887	-0.034327	0.068895	0.172606	0.110407	1.000000	-0.131964	0.195767	0.099832	0.180224	0.159859	0.125560	0.175071	0.093800
Spices	0.191468	-0.046152	0.044533	-0.083795	0.185123	0.252876	-0.084048	0.107548	0.001831	-0.023079	0.148281	-0.131964	1.000000	0.061399	0.272312	0.196014	0.199024	0.217365	0.141125	-0.014847
Malty	0.162060	0.300041	0.288219	-0.082085	0.754228	0.270106	0.565570	0.471032	-0.303266	-0.028241	-0.196890	0.195767	0.061399	1.000000	0.205665	0.288089	0.231032	0.236624	0.212990	0.006598
review_aroma	0.416197	0.343730	0.369456	0.068579	0.360234	0.242989	0.241457	0.356514	0.289320	-0.088210	0.386676	0.099832	0.272312	0.205665	1.000000	0.855629	0.907433	0.936670	0.870504	0.174033
review_appearance	0.349139	0.380908	0.379655	0.070948	0.421331	0.184693	0.331854	0.332311	0.204817	-0.081388	0.288594	0.180224	0.196014	0.288089	0.855629	1.000000	0.869978	0.850693	0.813444	0.191213
review_palate	0.392517	0.352723	0.355608	0.106950	0.345270	0.205569	0.257267	0.315767	0.245169	-0.038413	0.324906	0.159859	0.199024	0.231032	0.907433	0.869978	1.000000	0.946520	0.918154	0.185631
review_taste	0.381200	0.339885	0.346557	0.092021	0.365055	0.206701	0.253373	0.322277	0.244314	-0.057479	0.319690	0.125560	0.217365	0.236624	0.936670	0.850693	0.946520	1.000000	0.937102	0.190588
review_overall	0.251838	0.289972	0.271969	0.159788	0.313824	0.069789	0.260874	0.207293	0.213449	-0.008199	0.261504	0.175071	0.141125	0.212990	0.870504	0.813444	0.918154	0.937102	1.000000	0.176511
number_of_reviews	0.142176	0.219822	0.221572	-0.050062	0.039057	0.048346	0.112026	0.048487	0.006642	-0.029680	0.106223	0.093800	-0.014847	0.006598	0.174033	0.191213	0.185631	0.190588	0.176511	1.000000

Looks like 'Fruits' and 'Sour' has a strong correlation I didn't notice before.. I will try to create a regression model to predict sourness:

target = 'Sour'
train_cols =  [a for a in df.columns if a != target]
X = df[train_cols]
y = df[target]

X_train, X_test, y_train, y_test = train_test_split(X,y, test_size=0.30, random_state=42)

## Compare with baseline model, using just mean values:

baseline = y_train.mean()

y_pred_train = [baseline]*len(X_train)  
y_pred_test = [baseline]*len(X_test)


# squared=False or else you will be getting squared values!
train_rmse = mean_squared_error(y_pred_train,y_train,squared=False)
test_rmse = mean_squared_error(y_pred_test,y_test,squared=False)


train_r2 = r2_score(y_pred_train,y_train)
test_r2 = r2_score(y_test,y_pred_test)
print('rmse train:',train_rmse,'rmse test:', test_rmse)
print("r2 train:", train_r2,'r2 test:', test_r2)

rmse train: 36.329438302026844 rmse test: 34.448542812443726
r2 train: -6.53547213717035e+30 r2 test: -0.00030434775618526366

So the r-squared to beat is.. actually pretty bad. Should be easy to improve on this model.

## I will try a linear regression using the 'Fruits' values to predict level of 'Sour':

reg = LinearRegression()

X_train_univariate = X_train['Fruits'].values.reshape(-1,1)
X_test_univariate = X_test['Fruits'].values.reshape(-1,1)

reg.fit(X_train_univariate, y_train)
print("regression parameters:", reg.coef_[0],reg.intercept_)

regression parameters: 0.8688020267269813 -0.142996627279544

train_rmse = mean_squared_error(reg.predict(X_train_univariate),y_train,squared=False)
test_rmse = mean_squared_error(reg.predict(X_test_univariate),y_test,squared=False)


train_r2 = r2_score(reg.predict(X_train_univariate),y_train)
test_r2 = r2_score(reg.predict(X_test_univariate),y_test)
print('rmse train:',train_rmse,'rmse test:', test_rmse)
print("r2 train:", train_r2,'r2 test:', test_r2)

rmse train: 22.677000979470822 rmse test: 20.7748016291049
r2 train: 0.36164592053369427 r2 test: 0.4198851938837118

The r-squared is positive! Okay, that is okay for only using one variable, and it is an improvement from the baseline.

I can improve on this model with more variables, but first let's see that trendline:

## Evaluating the trend line:
df.plot.scatter(x='Fruits',y='Sour',marker='o',c="g",alpha=0.5)
plt.scatter(x=df['Fruits'],y=reg.predict(df['Fruits'].values.reshape(-1,1)),marker='+',label='trend_line')

<matplotlib.collections.PathCollection at 0x7fa7e2b1ffd0>

Let's see if adding 'Astringency' can help predict level of 'Sour' (sourness):

reg_multiple = LinearRegression()

multiple_columns = ['Fruits','Astringency']

reg_multiple.fit(X_train[multiple_columns], y_train)
print("regression parameters:", reg.coef_,reg.intercept_)

regression parameters: [0.86880203] -0.142996627279544

## Check r-squared for this model:
train_rmse = mean_squared_error(reg_multiple.predict(X_train[multiple_columns]),y_train,squared=False)
test_rmse = mean_squared_error(reg_multiple.predict(X_test[multiple_columns]),y_test,squared=False)


train_r2 = r2_score(reg_multiple.predict(X_train[multiple_columns]),y_train)
test_r2 = r2_score(reg_multiple.predict(X_test[multiple_columns]),y_test)
print('rmse train:',train_rmse,'rmse test:', test_rmse)
print("r2 train:", train_r2,'r2 test:', test_r2)

rmse train: 19.393115002418288 rmse test: 17.88770762454073
r2 train: 0.6014846979040045 r2 test: 0.6571811036469168

Hey, 60-65%, that's an improvement from a negative R-squared.

Let's see if any of the other variables with less hints of correlation can improve the model:

## Can adding metric for sweetness improve the model?:
reg_multiple = LinearRegression()

multiple_columns = ['Fruits','Astringency','Sweet']

reg_multiple.fit(X_train[multiple_columns], y_train)
print("regression parameters:", reg.coef_,reg.intercept_)

regression parameters: [0.86880203] -0.142996627279544

## Check r-squared with sweetness added:
train_rmse = mean_squared_error(reg_multiple.predict(X_train[multiple_columns]),y_train,squared=False)
test_rmse = mean_squared_error(reg_multiple.predict(X_test[multiple_columns]),y_test,squared=False)


train_r2 = r2_score(reg_multiple.predict(X_train[multiple_columns]),y_train)
test_r2 = r2_score(reg_multiple.predict(X_test[multiple_columns]),y_test)
print('rmse train:',train_rmse,'rmse test:', test_rmse)
print("r2 train:", train_r2,'r2 test:', test_r2)

rmse train: 19.22033019599842 rmse test: 17.791411883218753
r2 train: 0.6113022158064989 r2 test: 0.6617225875257344

Not much of an imrovement there. Let's try adding 'Malty', Astringency', and 'review_appearance':

#Adding 'Astringency', 'Malty' and 'review_appearance':
reg_multiple = LinearRegression()

multiple_columns = ['Fruits','Astringency','Malty','review_appearance','Sweet']

reg_multiple.fit(X_train[multiple_columns], y_train)
print("regression parameters:", reg.coef_,reg.intercept_)

regression parameters: [0.86880203] -0.142996627279544

train_rmse = mean_squared_error(reg_multiple.predict(X_train[multiple_columns]),y_train,squared=False)
test_rmse = mean_squared_error(reg_multiple.predict(X_test[multiple_columns]),y_test,squared=False)


train_r2 = r2_score(reg_multiple.predict(X_train[multiple_columns]),y_train)
test_r2 = r2_score(reg_multiple.predict(X_test[multiple_columns]),y_test)
print('rmse train:',train_rmse,'rmse test:', test_rmse)
print("r2 train:", train_r2,'r2 test:', test_r2)

rmse train: 18.456619232390146 rmse test: 17.163116228854356
r2 train: 0.6521105993052694 r2 test: 0.6929781616244106

This seems about the most accurate prediction model I will get ( ~ 65-69% correlation) for 'Sour' with the LinearRegression without changing variables, or by using other tuning.